Beyond Self-attention: External Attention using Two Linear Layers for Visual Tasks

TL;DR

This paper introduces external attention, a novel mechanism using shared memories and simple linear layers, offering a more efficient alternative to self-attention for various visual tasks with comparable or better performance.

Contribution

The paper proposes external attention, a new attention mechanism with linear complexity, replacing self-attention in architectures and enabling an all-MLP model for image classification.

Findings

External attention achieves comparable or superior results to self-attention.

It significantly reduces computational and memory costs.

Effective across multiple visual tasks including classification and segmentation.

Abstract

Attention mechanisms, especially self-attention, have played an increasingly important role in deep feature representation for visual tasks. Self-attention updates the feature at each position by computing a weighted sum of features using pair-wise affinities across all positions to capture the long-range dependency within a single sample. However, self-attention has quadratic complexity and ignores potential correlation between different samples. This paper proposes a novel attention mechanism which we call external attention, based on two external, small, learnable, shared memories, which can be implemented easily by simply using two cascaded linear layers and two normalization layers; it conveniently replaces self-attention in existing popular architectures. External attention has linear complexity and implicitly considers the correlations between all data samples. We further incorporate the multi-head mechanism into external attention to provide an all-MLP architecture, external attention MLP (EAMLP), for image classification. Extensive experiments on image classification, object detection, semantic segmentation, instance segmentation, image generation, and point cloud analysis reveal that our method provides results comparable or superior to the self-attention mechanism and some of its variants, with much lower computational and memory costs.